Learning from Hierarchical Structure of Knowledge Graph for Recommendation

Knowledge graphs (KGs) can help enhance recommendation, especially for the data-sparsity scenario with limited user-item interaction data. Due to strong power of representation learning graph neural networks (GNNs), recent works KG-based recommendation deploy GNN models learn from both knowledge and bipartite graph. However, these have not well considered hierarchical structure graph, leading sub-optimal results. Despite benefit structure, leveraging it is challenging since always partly-observed. In this work, we first propose reveal unknown structures a supervised signal detection method then exploit disentangling learning. We conduct experiments on two large-scale datasets, which results verify superiority rationality proposed method. Further ablation study respect key model designs demonstrated effectiveness our model. The code available at https://github.com/tsinghua-fib-lab/HIKE.